ArXiv versus AI slop: will capping preprint submissions work?

In a bid to tackle the deluge of manuscripts it now receives, the preprint server arXiv has announced a cap of two papers per month for each submitting researcher, which began on 1 October.

The move follows a steep rise in submissions driven mainly by the slop generated by artificial-intelligence tools, says Steinn Sigurðsson, an astrophysicist at Pennsylvania State University in University Park and scientific director of arXiv. In September, 40,000 preprints were uploaded to arXiv, up from about 30,000 between March and July.

Closing the floodgates

The cap is the latest measure introduced by arXiv to prevent its nearly 300 voluntary moderators from being completely overwhelmed by submissions, says Sigurðsson. They are now collectively checking about 1,000 manuscripts per day. This step is “holding up too many legitimate papers,” Sigurðsson says, noting that nearly 4,000 submissions are currently waiting to be posted on the site.

Some see the limit as entirely reasonable. Siddhesh Zādéy, an epidemiologist at Columbia University in New York City, says that it is not “stunting anyone's productivity”, because posting two preprint articles per month already requires substantial work.

Surge in fake citations uncovered by audit of 2.5 million biomedical-science papers

Others think that arXiv’s new policy should be stricter. Victor Galitski, a theoretical physicist at the University of Maryland in College Park, says that the limit can be circumvented through co-authors taking turns to submit studies.

Another preprint server, the social-sciences repository SocArXiv, now also restricts submissions to one article per author per week. “Sometimes there are good reasons to upload three papers at once, such as dissertation chapters,” notes Philip Cohen, a sociologist and demographer at the University of Maryland, who is director of SocArXiv. But most of the time, several submissions in rapid succession are “a good tip-off for AI slop”, he says.

Other venues are dealing with a rapid surge in submissions, and some major AI conferences and journals have turned to similar submission quotas.

Clean up the literature

Some say that the issue of AI slop can’t be solved by rejecting more articles. “If arXiv turns those papers away, they’ll simply go somewhere else and become someone else’s problem,” says Natalie Khalil, the founder of Reviewer3, a platform run from San Francisco, California, which uses AI to help researchers conduct peer review.